Tuesday, 15 April 2025

The Shape of AI to Come -- catching up with AI ART

AI ART is a big blur now, evolving so fast that I can barely keep up with all the new approaches and wrinkles -- and to think, I declared that AI ART was slowing down last November: 

WAN 2.1 VIDEO

WAN 2.1 Local Generations

Adam and I generated the AI videos above after "pulling" a Docker Image (28 GIGs) with an AI video generation program -- WAN 2.1 -- on April 12, 2025. More significantly, we generated the videos LOCALLY, on the supercomputer at Quelab, with 10 GIGs of GPU.  

I also uploaded an AI version of a portrait that I drew into the WAN 2.1 Docker Image, and generated the AI video below:

Realistic video on the right,
generated in WAN 2.1 from the image on the left

I had previously generated AI videos ONLINE in KREA-- using WAN 2.1 -- and added them to my YouTube videos:
MCP
The Shape of AI to Come

Recently I have been merely chatting with Blender, to generate 3D scenes using MCP:

I merely typed in prompts
to generate this 3D scene in Blender

"An intersection in a city"
generated by chatting to Blender 


Tim set up the Quelab computer so that Claude Desktop would talk to Blender directly, using MCP protocol, following the instructions from the YouTube below:

I even imported a GLB scene I had previously constructed from 3D objects -- that I generated in Google AI Studio -- and added trees and other elements to the 3D Blender scene, by merely having a conversation with Claude Desktop using MCP:

 
The original scene (top) --
I uploaded it to Blender and added the trees (bottom)
by merely chatting with Claude Desktop


NOTE: Generating 3D OpenSCAD code (to make 3D objects after uploading 2D images) is no longer free in Google AI Studio.  April 14, 2025 the free "Gemini 2.5 Pro Experimental 03-25" switched to the paid "Gemini 2.5 Pro Preview 03-25."  

MCP (Model Context Protocol) is a new standard for AI programs to talk to each other and other software, making it easier for us users to use several programs to create a single video or other project.


KREA

KREA let us train 5 free AI models for a week, after they revamped their website on March 31, 2025.  So I trained 6 AI models (one on the paid plan) on my drawings, and altered my drawings with those AI models, and was pleased with the result.  I also subscribed to the BASIC plan for a year:

My figure drawing from an ink pen,
transformed by a KREA AI model based on my drawings

KREA Blog Posts:

Download: Standing__A.SVG  (108.19 KB)

Download: Standing__B.SVG  (116.49 KB)

Colored with AI
using Google AI Studio


Download: Standing__C.SVG  (133.32 KB)

November 29, 2022
Autotracer
Download "221129__B.SVG"  (30.50 KB)
The original pen drawing


December 1, 2022
KREA
My abstract figure drawing with an ink pen,
transformed into a recognizable figure with KREA AI

December 1, 2022
Download: Sitting_A.SVG  (207.28 KB)


When I do a reverse search on my AI images in Google Lens, I do not really find anything similar to them.  Am I generating ORIGINAL AI ART?  After all, I'm uploading my own figure drawings, and altering them with an AI model trained on my figure drawings.

However my AI results look a lot like the drawings generated in Doodly.

 

Posted by at

No comments:

Post a Comment

Note: only a member of this blog may post a comment.

Subscribe to: Post Comments (Atom)